Ontology-Driven Conceptual Modelling: A 4D Extensionalist Approach Sergio de Cesare Professor of Digital Business Centre for Digital Business Research Outline Problem statement and conceptual modelling Ontology: a philosophical stance Metaphysical choices Modelling with different metaphysical choices Semantic improvement of legacy data Conclusions 2 How many countries? 1, 4 or 5? What happens if a product is sold in Scotland? Is it also sold in the U.K.? What if Scotland ceases to be part of the UK? Would it be the same UK or a different country? These questions exemplify some of the problems underlying the development and integration of modern information systems. 3 style.visibility style.visibility style.visibility style.visibility Traditional definitions of conceptual modelling In the design, development, evolution and integration of information systems, conceptual modelling plays a key role In a nutshell, its purpose is to represent the problem domain (or domain of discourse) Some popular definitions include: Conceptual modelling can be described as the activity of representing aspects of the physical and social world for the purpose of communication, learning and problem solving among human users (Mylopoulos, 1992) Conceptual modelling in information systems development is the creation of an enterprise model for the purpose of designing the information system (Wand et al., 1995) Are these definitions sufficient for the modern context of enterprise systems development and integration? The modern enterprise context is one of flux, with continuously evolving requirements (as in the country example) Machine-to-machine interaction (or interoperability) is becoming the norm, with data being exchanged between software systems (this represents a difficult and costly problem) style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility The data integration problem 5 Unstructured datasets Structured datasets style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility Challenges of Data Integration Increasing amounts of data in multiple syntactic formats (CSV, JSON, XML, RDF, text, etc.) Data is not clean at the onset Open and private datasets are increasingly available Continuous flow of incoming data in the order of Terabytes, Petabytes or Exabytes For unstructured data, it is a problem of Natural Language Processing (NLP): NLP and Artificial Intelligence have progressed significantly, but they are still not effective in transforming text into reliable structured forms (and models) For structured data (the main focus of this talk), challenges include: Syntactic, structural and semantic patterns for integration Semantics plays a key role AI is unable to accurately mine the semantics 6 It’s just semantics … In everyday life people tend to minimise the importance of semantics “It’s just semantics” Stephen Pinker in the “Stuff of Thought” writes: “But I would like to explore a lesser-known debate triggered by 9/11. Exactly how many events took place in New York on that morning in September?” The difference was worth $3.5 billion 7 style.visibility style.visibility style.visibility style.visibility style.visibility What is semantics? Semantics is meaning. Meaning is Semantics. This definition does not go far Semantics as reference or mapping Given a model what do the symbols in the model map/refer to in the real world? Semantics as sense and reference ( Frege ) Sense relationships such as the relationship between the UK and England Relationships as first-class objects 8 Sense and Reference From Partridge (1996) 9 What is this a picture of? From Partridge (1996) 10 Modelling Paradigms Generally speaking models of systems are created according to a particular ‘ worldview ’ or way of seeing the world This ‘ worldview ’ or perspective is often known as paradigm (Thomas Kuhn - The Structure of Scientific Revolutions (1962)) Different paradigms lead to different interpretations of the world and, therefore, different types of models. For example: Evolutionary vs. Creationist paradigms Ptolemaic (geocentric) vs. Copernican (heliocentric) paradigms 11 What is Ontology? In Philosophy Lowe (1995) defines ontology as “ the set of things whose existence is acknowledged by a particular theory or system of thought ” Ontology is “the set of things”. When modelling reality, one might think that one should recognise all those things (or objects) that can be referred to. However, identifying objects should not be a simple subjective listing. Therefore, one should adopt a “particular theory or system of thought”, that is, a theoretical basis that defines the set of ontological commitments made, in other words, what things the ontology commits to existing. A foundational ontology makes a set of metaphysical (or meta-ontological) choices, carving reality in specific ways and committing to specific top-level categories of objects Making the case for philosophical ontology: Data model real-world problem domains ( problem space ) Philosophical ontology provides a real-world grounding for the models produced The closer the underlying model of a data representation is to the real world, the easier it is to integrate datasets (even automatically between systems) in a consistent manner 12 Metaphysical Choices {5C22544A-7EE6-4342-B048-85BDC9FD1C3A} Choice 1 Choice 2 Related Topics Endurantism Perdurantism Existence. Change. Eternalism Presentism Existence. Change. Single Space-Time Continuum Separate Space and Time Continua Change. Modally Extended Modally Flat Modality/Possibility. Counterparts. First Order Universals Only Higher Order Universals Existence. Universals – Metaphysical Realism Universals – Nominalism Identity. Can two different universals have the same extension? Particulars – Extensional Identity Particulars - Coincident Identity. Incudes mereology. Materialism Non-Materialism (Abstract) Existence. Branching Time Linear Time Existence. Possibility. Partridge, C.: LADSEB-CNR - Technical Report 06/02 - Note: A Couple of Meta-Ontological Choices for Ontological Architectures. LADSEB CNR, Italy, (2002) An example of a metaphysical choice Endurantism vs perdurantism Two Schools of Thought: 3D Endurantism or 3D A three-dimensional object is wholly present at any given instant and persists by appearing in different spaces and different times While wholly present at all moments of its existence, an object preserves its identity via a set of essential attributes (for example, a person’s DNA) John and Mary are different because they have different properties (Leibnitz's identity of indiscernibles ) 15 style.visibility style.visibility style.visibility style.visibility Two Schools of Thought: 4D Perdurantism or 4D An individual thing has a four-dimensional extension (or extent) in the universe (i.e., the region of space-time that it occupies), and it is not therefore totally present at any given instant, but instead only partially present Identity is defined by the thing’s four-dimensional extension In its lifetime, an individual thing goes through states (or stages) Change is explained via successive dissimilar temporal parts John and Mary are different because they occupy different spatiotemporal extensions in the universe (more technically, they have different parts) 16 style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility Space-Time Diagram: Temporal Parts 17 Sider: http:// tedsider.org /papers/ temporal_parts.pdf Example Acme Ltd. is founded with paper production as its main business activity Acme registers a change in its SIC code due to a change in its primary business activity It then produces electronics Acme is ultimately dissolved 18 3D Interpretation Change is handled via change in membership 19 Paper Companies Electronics Companies Companies Acme Ltd. From t 1 to t 2 Paper Companies Electronics Companies Companies Acme Ltd. From t 2 to t 3 style.visibility style.visibility 4D interpretation Acme Ltd. is founded with paper production as its main business activity Acme registers a change in its SIC code due to a change in its primary business activity It now produces electronics Acme is ultimately dissolved 20 Acme Ltd. Incorporation Paper Production Temporal Part (State #1) Electronics Production Temporal Part (State #2) SIC Code Change Dissolution Space Time t 1 t 2 t 3 creation event (E1) dissolution event (E3) change event (E2) 4D interpretation 21 Companies Acme Ltd. Company States Electronics Production States Paper Production States State #1 State #2 Acme Ltd. Incorporation Paper Production Temporal Part (State #1) Electronics Production Temporal Part (State #2) SIC Code Change Dissolution t 1 t 2 t 3 temporalWholeParts creation event (E1) dissolution event (E3) change event (E2) (Acme, State #1) (Acme, State #2) (Acme, E1) (Acme, E2) (Acme, E3) style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility Basis for the 4D Approach Model all business things in a similar manner Overcome the traditional distinction between static and dynamic models Adopt an upper-level ontology 4D models Empirically discover patterns from organisational data (not simply based on consensus or opinion) Continuously test models/patterns Create a repository of reusable, tested patterns 22 Representing Processes A simple example A type of gardening service Lawn mowing service Simplified example to illustrate the modelling approach 24 Process of Lawn Mowing (traditional representation) Process Type Process instance 25 style.visibility style.visibility style.visibility style.visibility What things are involved? 26 … but these things exist before and after the process as well Time Space John Lawn Mower Joe Lawn style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility So what exactly takes part in the process and constitutes a process? 27 Time Space John Lawn Mower Joe t 2 t 3 t 4 Lawn D C F E t 1 t 5 H A B G J I t 6 A, B C, D, E, F G, H I, J style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility style.visibility What is a business process? 28 A business process has temporal parts that intersect with other objects Time t 2 t 3 t 4 t 1 t 5 Time Space John Lawn Mower Joe t 3 t 4 Lawn D C F E t 5 H A B G J I t 6 P = Lawn Mowing Service A process is the mereological sum of temporal parts 29 Time D C F E H A B G J I P is composed of temporal parts A to I. P intersects with other objects. Space Benefits Precise definition of business process Real-world grounding Processes, activities, etc. not simply defined as abstract things It is possible to know what we are referring to We can count the things involved Two modellers should be able to model the same phenomenon in the same way Precisely identify commonalities and differences Patterns of process types and their variations 30 Foundational ontology A foundational ontology can be defined as an ontology that ‘‘defines a range of top-level domain-independent ontological categories, which form a general foundation for more elaborated domain-specific ontologies’’ ( Guizzardi and Wagner 2004). A foundational ontology defines the kinds of things that exist (aka categorical ontology) In our research, we adopt the Business Object Reference Ontology (BORO) A realist ontology (i.e. there is an objective physical real world) Adopts a 4D or perdurantist theory of existence Extensionalism (to define identity) Possible worlds (counterparts) Three top-level categories: Elements, Types, Tuples 31 A re-engineering example Countries An apparently simple example: Countries The following example is drawn from a real project related top the re-engineering of different financial systems The systems had relational-type databases (table-based) All financial systems have some form of country table Countries can be classified in different ways and each system used a different classification These different classifications needed to be integrated 33 Example of the Legacy Country Table 34 Two Example Implementations of Country Table 35 Competency Questions CQ1: Can a country instantiate multiple classifications of countries? CQ2: Can multiple country classification systems be represented? CQ3: Can both nesting (e.g., United Kingdom) and nested countries (e.g., England, Northern Ireland, Scotland and Wales) be represented? 36 Content Interpretation 37 Identifying Subtypes of Countries 38 Adding a new Country Subtype 39 CQ1: Can a country instantiate multiple classifications of countries? CQ1 SATISFIED Being able to represent Country Classifications 40 Explicit Representation of Country Classification Systems and Their Groupings 41 CQ2: Can multiple country classification systems be represented? CQ2 SATISFIED Whole-Part Relationships between Countries 42 Final Semantically Improved BORO Model for Countries 43 CQ3: Can both nesting (e.g., United Kingdom) and nested countries (e.g., England, Northern Ireland, Scotland and Wales) be represented? CQ3 SATISFIED This model now integrates all classifications of countries used by the different systems and, in addition, allows for the representation of nested countries. Semantic Improvements 44 {284E427A-3D55-4303-BF80-6455036E1DE7} Dimensions Improvements Objectivity The new model is not dependent on a specific system implementation or an organization’s perspective since multiple classifications of countries are now possible within the same enterprise system. Generality The new model is more general than the legacy model since it allows for multiple country classifications and mereological relationships between countries. Precision The new model is a more accurate representation of countries since it is able to represent country whole-part relationships. Explanatory power The new model provides a set-theoretic definition of country classifications, and it is capable of explaining what country classification systems are, as well as providing the support to classify the classifications themselves. Simplicity At the type level the new model contains a small set of objects: two types (countries and its powertype ), and two tuple types (country whole-parts and the powertype instance relation). At the individual element level, the new model represents each country once (via multiple instantiation) rather than multiple times for each implemented classification, as in the legacy system. Precision The new model is a more accurate representation of countries since it is able to represent country whole-part relationships. Adapted from Kuhn (1977) Objectivity, value judgment, and theory choice. Conclusions Mainstream information systems engineering and data modelling are characterised by difficult and costly problems A possible cause is the absence of sound theories underpinning these activities Philosophical ontology can assist in grounding and mapping data to reality via sound ontological theories Government seems to be starting to adopt foundational ontologies like BORO (e.g., the Information Exchange Standard (IES)) Educating the next generation is crucial